Integrating Wearable Sensor Signal Processing with Unsupervised Learning Methods for Tremor Classification in...

Area of Science:

• Biomedical Engineering
• Neurology
• Data Science

Background:

• Tremor is a common Parkinson's disease (PD) symptom, often assessed via subjective clinical scales.
• Wearable sensors offer potential for objective, continuous tremor monitoring.

Purpose of the Study:

• To evaluate unsupervised learning for classifying and assessing tremor severity using wearable accelerometer data.
• To explore the utility of k-means clustering for PD tremor analysis.

Main Methods:

• Analysis of resting tremor signals from 24 participants (13 PD patients, 11 controls) using accelerometer data.
• Application of k-means clustering for classifying tremor vs. non-tremor states and tremor severity levels.

Main Results:

• K-means achieved 76% accuracy in classifying tremor versus non-tremor states.
• Multiclass tremor severity classification accuracy was 57.1%, and binary classification (severe vs. mild) was 71.4%.

Conclusions:

• Unsupervised learning demonstrates potential for scalable, objective tremor analysis in PD.
• Wearable sensor integration could enhance monitoring and clinical assessments, but further algorithm development is needed for improved severity classification.